Known methods of electrically connecting a semiconductor device to a liquid crystal display device, a circuit substrate or a TCP include a wireless bonding method wherein a bump electrode is provided on an electrode terminal of a semiconductor device, and the semiconductor device is bonded directly to the liquid crystal display device, the circuit substrate or the TCP using the bump electrode.
A concrete example will be given through the case of the liquid crystal display device of a chip-on-glass (hereinafter referred to as COG) method wherein a liquid crystal driver chip as a semiconductor device is directly face-down-bonded to a liquid crystal panel. The described COG method is classified into two types: (i) a paste COG method wherein the liquid crystal driver chip and the liquid crystal panel are connected by means of an electrically conductive paste; and (ii) an ACF-COG method wherein the liquid crystal driver chip and the liquid crystal panel are connected by means of an anisotropic electrically conductive film (hereinafter referred to as ACF).
As shown in FIG. 13(a), a liquid crystal driver chip A′ includes a semiconductor base 101 whereon an insulating film 102, an electrode pad 103, and a protective film 104 with an opening are laminated in this order. On the opening, a bump electrode 106 is formed via a barrier metal 105. On the other hand, a liquid crystal panel B′ on which the described liquid crystal drive chip A′ is to be bonded includes a glass substrate 109 whereon a conductive pattern 110 and a protective film 111 are laminated in this order.
In the paste COG method, after a conductive paste 112 is applied to the bump electrode 106 of the liquid crystal driver chip A′, the liquid crystal driver chip A′ is face-down-bonded to the liquid crystal panel B′. Then, the conductive paste 112 is cured, thereby connecting the bump electrode 106 and input and output terminals of the conductive pattern 110n. 
On the other hand, in the ACF-COG method, as shown in FIG. 13(b), an ACF composed of a binder resin 107 and conductive particles 108 is enclosed in a spacing formed between the liquid crystal driver chip A′ and the liquid crystal panel B′, thereby connecting the bump electrode 106 and the input and output terminals of the conductive pattern 110 via the conductive particles 108.
However, the described COG method wherein the liquid crystal driver chip is directly face-down-bonded to the liquid crystal panel has a drawback in that a mounting inferior occurs when mounting the panel due to a bump inferior such as a bump defect, irregularity in bump height, etc.
Moreover, in the COG method, as the bump electrode is bonded to a hard material such as glass, etc., irregularity in height of bumps within the liquid crystal driver chip would cause a problem. For example, in the case of the ACF-COG method, the conductive particles in the ACF has an average particle diameter in a range of from 3 to 5 μm, and thus if a gap between heights of the adjoining bump electrodes is larger than the diameter of the conductive particle, a connection inferior occurs. Also, in the case of the paste COG method, if a gap between heights of the adjoining bump electrodes is larger than the thickness of the paste to be applied, a connection inferior occurs.
When adopting the COG method, in order to prevent an increase in a contact resistance, or a contact inferior due to an unexpected bump defect, a multiple port structure is generally adopted for the power source terminal and the input terminal. For the output terminal, however, in consideration of a space required, etc., the multiple port structure is not adopted.
Therefore, even when only one of the bump electrodes on the output terminal is defective, or significantly lower than the adjoining bump electrode (by not less than a conductive particle diameter), a panel display defect such as a line defect, etc., occurs, or fixing of the defective bump electrode is required or wasted otherwise.
The described problem occurs not only in the liquid crystal display device of the COG method but also in the circuit substrate wherein the semiconductor chip is face-down-bonded to the substrate main body such as a print substrate, a ceramic substrate, etc. Such problem occurs because the substrate main body to which the bump electrode is bonded is made of a hard material.
The TCP has advantageous features over other face-down-bonding method in that (i) the inner lead is flexible, and (ii) an eutectic crystal is generated by the bump electrode and Sn plated onto the inner lead, and the inner lead is inserted into the bump electrode. However, even for the described TCP, for example, if a bump defect occurs, or a significant gap in bump height is generated, a connection inferior cannot be avoided.
In order to counteract the described problem, as shown in FIG. 14, Japanese Unexamined Utility Model Publication No. 56136/1991 (Jitsukaihei 3-56136) discloses a bonding bump wherein four divided gold bumps (bump electrodes) 206a, 206b, 206c and 206d are formed on a square connection terminal (electrode pad) 206. According to the described arrangement, a semiconductor chip is bonded to a substrate terminal by means of a curing resin around the metal bump. Therefore, by dividing the metal bump into four, an occurrence of a connection inferior due to a residual resin remaining between (a) the metal bumps 206a, 206b, 206c and 206d, and (b) the substrate terminal can be suppressed.
However, the described arrangement of Japanese Unexamined Utility Model Publication No. 56136/1991 has a drawback in that as the bump electrode is divided bidirectionally along a column and a row, narrowing of an electrode pad pitch, i.e., a wiring pitch is not possible.
Moreover, the arrangement wherein the bump electrode composed of a transferred bump substrate is directly bonded to the semiconductor chip like the case of the above Gazette cannot be applied when the electrode pad pitch is not more than 100 μm. Recently, as an electrode pad pitch of from 50 to 80 μm has been generally adopted to meet a demand for a miniaturization of a semiconductor chip, the described method cannot be used in practical applications.
The described mechanism is the same as the following mechanism. The generally used transfer bump adopts an inner lead bonding method wherein the bump electrode is transferred to a leading end of the inner lead of the tape carrier, and the bump electrode thus transferred is inner-lead-bonded to the electrode pad of the semiconductor chip. Although this method has an advantageous feature in that a wafer bump process can be omitted, as a mechanical connection is repeated twice, it is not practical to use the method for the electrode pad pitch of not more than 100 μm in view of precision. Therefore, in consideration of the facts that the mechanical connection is required, and the described transferring of the bump electrode can be performed only chip by chip, a wafer bump method wherein the bump electrode is formed on the electrode pad of the semiconductor chip in the wafer bump process is superior to the transfer bump method in view of both precision and mass production.
As to the divided bump electrodes, Japanese Unexamined Patent Publication No. 13418/1993 (Tokukaihei 5-13418) and Japanese Unexamined Patent Publication No. 58112/1995 (Tokukaihei 7-58112) disclose four divided bump electrodes for the purpose of suppressing a generation of a thermal stress.
However, neither of described Gazettes teach two divided bump electrodes or the mounting structure which offers a lower rate of an inferior such as a bump defect, etc. These Gazettes also fail to refer to the wiring pitch. Especially, Japanese Unexamined Patent Publication No. 13418/1993 (Tokukaihei 5-13418) discloses the arrangement wherein a solder layer is formed on the bump electrode. According to the described arrangement, even when only one of the bump electrodes is defective, the solder layer cannot be formed in a shape as desired, and an occurrence of a bump inferior cannot be suppressed. On the other hand, Japanese Unexamined Patent Publication No. 58112/1995 (Tokukaihei 7-58112) discloses the diode element only, and does not refer to the concept of the wiring pitch.